What is the topology of Imageline's Unique Poizone's HP filter (Sounds like ring modulation in it at high rez)

[Touch The Universe]

It is unique. I want to recreate it. It sounds like the resonance generates self oscilation which itself becomes a frequency or ring modulation to the saw wave or something. HP has the fullest affect, BP has it a bit, LP has it very little.

[Touch The Universe]

Is it ring modulation in the filter path? I haven't hear this sound anywhere else. Is it a mystery? no one has any ideas? I thought kvr was the land of genuises and "know it alls"

[Touch The Universe]

wt. 2

[Touch The Universe]

wt

[Tiles]

Resonance feedback seems to be a factor. But i don't know enough about synths and effects to be able to judge.

chatgpt thinks

What you're hearing is probably a combination of:

Resonance feedback distortion
Phase interaction inside the HP topology
Harmonic emphasis moving across oscillator partials
Possible coefficient instability or nonlinearity at high cutoff/Q

[Touch The Universe]

Yeah, i had it give it a go and the filter it gave me to try wasn't close and we went deep on it. It's a mystery.

[Tiles]

Try Claude, or another solution like Deepseek. It might have a different approach.

[BertKoor]

Try Claude, or another solution like Deepseek. It might have a different approach.

Futile imho. These are not the Thinking Machines they want you to believe, but merely stochastic parrots. They cannot really solve hard problems or reason about things. They merely echo from past online conversations. That should be common knowledge by now.

Although if you ask the right questions, eg "what tests can be performed to analyse nonlinear filter behaviour" then you might get further, at least in determining how close you have mimicked behaviour.

[Tiles]

Yeah, but i ran into the same problem, and asked Claude then, and Deepseek, and ChatGPT, and then i got the answer i was looking for. The one parrot sometimes knows more than the other

Going forwards and backwards between different AI solutions sometimes also leads to the right question. That's indeed an important bit, the art of asking.

[Touch The Universe]

Yeah, but i ran into the same problem, and asked Claude then, and Deepseek, and ChatGPT, and then i got the answer i was looking for. The one parrot sometimes knows more than the other

Going forwards and backwards between different AI solutions sometimes also leads to the right question. That's indeed an important bit, the art of asking.

I got it running another task, but this is worth a try. can you help make this an filter? request real dsp code for juce for an hp filter to mimic it and i will build it and showcase it, especially if its good. i tried but it's pretty not right .. You can ask it to prepare a .md file which is like learning, notes, studies, or whatever it compiles and share that file across the agents and it can have implementatin ideas on how to implement the code. That might sound complicated but see if it can spit out useful code to try, even a various amount of them, trial and error.

I wonder if it's only a filter or if we somehow need to have an osc/filter combo.

It requested filter sweeps and 4 second passes of filter at 17% , then 18% etd, and spetracl info and graphic.

It spit out graphs but i need to find them

[Tiles]

Sorry buddy i just realized i must have phrased it incorrectly. It is not the same problem, i haven't solved the riddle. It was a similar problem regarding getting stuck. First AI gave me nonsense, second came closer, third then solved my problem for me. That's what i meant. But let me pick the information that we have here and throw it at Deepseek. EDIT Claude ...

[Tiles]

Well, not so far away, what a coincidence. I have just built this filter here, and it uses the SFV algorithm under the hood.

What Claude says:

What Does the PoiZone Filter Actually Do?
PoiZone describes its filter as "warm, analog-modelled with self-oscillation" — it can become "gritty and grungy" when resonance is introduced, including "carefully crafted light distortion." zZounds
Your observation — HP has the strongest effect, BP medium, LP the least — is the crucial clue. This is not normal Moog ladder filter behavior (which is purely LP-based). It points to a State Variable Filter (SVF) topology combined with nonlinearities. Here's what's most likely happening:

The Hypothesis: SVF + Nonlinear Feedback + Tanh Waveshaper
1. State Variable Filter Topology
An SVF gives you LP, BP, and HP simultaneously from a single integrator network. That alone already explains the multimode behavior.
Input → [+] → Integrator1 → [HP out]
↑ ↓
feedback Integrator2 → [BP out]
↓
Integrator3 → [LP out]
2. Where Does the "Ring Mod" Character Come From?
This is not a separate ring modulator — it's the behavior of a self-oscillating SVF with a waveshaper inside the feedback loop. The physics:

At high resonance, the filter begins to oscillate nearly sinusoidally (self-oscillation)
When a tanh or soft-clip waveshaper sits inside the feedback loop, intermodulation products are generated between the incoming saw wave and the self-oscillating frequency
This sounds exactly like ring modulation / FM, but it's a byproduct of nonlinear feedback — not a separate module

3. Why HP > BP > LP?
With a Moog-style filter, self-oscillation sounds like a nearly perfect sinusoid — but waveshapers inside the feedback loop generate additional harmonics, which is why oversampling or other measures are needed. Willpirkle
In the SVF: the HP output has higher sensitivity to phase accumulation in the loop. The intermodulation between the incoming signal and the resonance sine is most audible at the HP output, because the HP path suppresses the resonant signal less than LP does.

The Reconstruction Approach in JUCE
Step 1: Nonlinear SVF (the core)
The most well-known model for this is the Cytomic/Mystran SVF (Andrew Simper) — a delay-free loop SVF with tanh saturation:
cppfloat processSample(float x)
{
float hp = (x - 2.0f * R * v1 - g * v1 - v2)
/ (1.0f + g * (2.0f * R + g));
float bp = g * hp + v1;
float lp = g * bp + v2;

// Nonlinear state update — tanh inside the loop
v1 = tanh(g * hp + v1); // <-- this is the heart of the sound
v2 = tanh(g * bp + v2);

return hp; // or lp / bp depending on mode
}
Where g = tan(π * fc / sampleRate) and R = 1 / (2 * resonance).
Step 2: Oversampling (critical!)
Waveshapers inside the loop generate aliasing, which is why oversampling or other countermeasures are necessary. Use at least 2x oversampling, preferably 4x. In JUCE this is straightforward with dsp::Oversampling. Willpirkle
Step 3: Drive Before the Filter
The "grungy" character intensifies further if you add a drive parameter (soft-clip) before the SVF:
cppfloat driven = tanh(drive * input);
float filtered = svf.processSample(driven);
Step 4: HP/BP/LP Output Blend
Since LP, BP, and HP are all available, you can expose a mode blend parameter:
cppfloat out = lpGain * lp + bpGain * bp + hpGain * hp;

JUCE Architecture
PluginProcessor
├── Oversampler (dsp::Oversampling, 2x or 4x)
├── NonlinearSVF (custom class)
│ ├── g, R, v1, v2 as state variables
│ ├── tanh() in state update
│ └── LP / BP / HP outputs
└── Parameters: cutoff, resonance, drive, mode (LP/BP/HP/blend)
The critical difference from the standard JUCE dsp::StateVariableTPTFilter: the JUCE built-in has no nonlinearities inside the loop. You need to implement the SVF manually.

Why Does It Sound Like Ring Modulation?
When self-oscillation runs at frequency f_res and your saw comes in at f_note, the nonlinear feedback generates products at f_note ± f_res, f_note ± 2·f_res, etc. — identical to the sonic signature of a ring modulator, but emerging entirely from within the filter physics. That's why you haven't heard this sound anywhere else: it's a side effect of the nonlinearity, not a dedicated module.

Want me to write out the complete JUCE C++ implementation of the nonlinear SVF?

Moog ladder is still possible though. I would throw this post on Claude and provide the code parts you have.

EDIT, now claude thinks it could also be a ...

Huovilainen Ladder (der wahrscheinlichste Kandidat)
cpp// Jede Stage hat einen Tanh-Eingang
float processSample(float x)
{
float feedback = resonance * 4.0f * tanh(output); // nichtlineare FB

float s0 = tanh(x - feedback);
float s1 = tanh(stage[0] + g * (s0 - stage[0]));
float s2 = tanh(stage[1] + g * (s1 - stage[1]));
float s3 = tanh(stage[2] + g * (s2 - stage[2]));
float s4 = tanh(stage[3] + g * (s3 - stage[3]));

// LP = s4
// HP = s0 - s4 (oder x - s4)
// BP = s2 - s4

stage[0]=s1; stage[1]=s2; stage[2]=s3; stage[3]=s4;
output = s4;
}

[Touch The Universe]

poizone_cutoff_sweep_peak_timeline.png

poizone_exact_notes_measured_pulses.png

poizone2_pulse_timeline.png

poizone2_env_20hz.png

poizone2_env_7hz.png

poizone2_env_34hz.png

poizone3_section_1_pulse.png

I’m trying to understand/recreate a particular filter behavior in Image-Line PoiZone.

The sound I’m hearing is on a saw patch with very high resonance, especially in the HP filter mode. It sounds almost like ring modulation, tremolo, or a low-frequency gated pulse, but I’m fairly sure it is not the oscillator ring mod.

What I tested:

- Ring mode is off.
- Sync is off.
- Only one oscillator is active.
- Changing the pitch of the second oscillator does not change the effect.
- The effect is strongest in HP mode.
- BP has some of it.
- LP mostly does not.
- With filter keytrack at 0%, different notes produce different pulsing/beating speeds.
- When I increase keytrack, the pulsing becomes more consistent across notes.

So my current theory is that this is not true oscillator ring modulation. It seems more like the high-resonance HP filter is creating a strong resonant tone/peak, almost like a hidden sine tone, and that resonant peak is beating against the harmonics of the saw wave.

In other words:

saw harmonics + smooth high-Q HP/BP resonance = pseudo-ring-mod / gated beating effect

The part that makes PoiZone special is that the resonance is very strong but still smooth and buttery. It does not sound like harsh distortion or a sharp digital whistle. I tried recreating it with a dirty resonant HP/SVF style filter, but that got too edgy and distorted. PoiZone’s version feels cleaner, softer, and more refined.

What I’m trying to figure out:

1. What kind of filter topology could create this?
2. Is PoiZone likely using an SVF, biquad, ladder-derived HP, OTA-style filter, or something custom?
3. Is the HP filter likely 12 dB, 24 dB, or something else?
4. Is the resonance peak self-oscillating, or just very close to self-oscillation?
5. Could the smoothness come from resonance gain compensation, internal smoothing, saturation control, or a particular HP/BP mixing structure?
6. Why would HP show this effect strongly, BP somewhat, and LP barely at all?
7. What would be the best way to recreate this kind of smooth high-resonance HP beating in a synth filter?

The goal is not to add an LFO tremolo or fake ring mod on top. I’m trying to reproduce the natural pulsing/beating that seems to come from the filter resonance interacting with the saw harmonics.

Any ideas on what filter design or DSP structure might create this behavior?

[Touch The Universe]

Well, not so far away, what a coincidence. I have just built this filter here, and it uses the SFV algorithm under the hood.

What Claude says:

What Does the PoiZone Filter Actually Do?
PoiZone describes its filter as "warm, analog-modelled with self-oscillation" — it can become "gritty and grungy" when resonance is introduced, including "carefully crafted light distortion." zZounds
Your observation — HP has the strongest effect, BP medium, LP the least — is the crucial clue. This is not normal Moog ladder filter behavior (which is purely LP-based). It points to a State Variable Filter (SVF) topology combined with nonlinearities. Here's what's most likely happening:

The Hypothesis: SVF + Nonlinear Feedback + Tanh Waveshaper
1. State Variable Filter Topology
An SVF gives you LP, BP, and HP simultaneously from a single integrator network. That alone already explains the multimode behavior.
Input → [+] → Integrator1 → [HP out]
↑ ↓
feedback Integrator2 → [BP out]
↓
Integrator3 → [LP out]
2. Where Does the "Ring Mod" Character Come From?
This is not a separate ring modulator — it's the behavior of a self-oscillating SVF with a waveshaper inside the feedback loop. The physics:

At high resonance, the filter begins to oscillate nearly sinusoidally (self-oscillation)
When a tanh or soft-clip waveshaper sits inside the feedback loop, intermodulation products are generated between the incoming saw wave and the self-oscillating frequency
This sounds exactly like ring modulation / FM, but it's a byproduct of nonlinear feedback — not a separate module

3. Why HP > BP > LP?
With a Moog-style filter, self-oscillation sounds like a nearly perfect sinusoid — but waveshapers inside the feedback loop generate additional harmonics, which is why oversampling or other measures are needed. Willpirkle
In the SVF: the HP output has higher sensitivity to phase accumulation in the loop. The intermodulation between the incoming signal and the resonance sine is most audible at the HP output, because the HP path suppresses the resonant signal less than LP does.

The Reconstruction Approach in JUCE
Step 1: Nonlinear SVF (the core)
The most well-known model for this is the Cytomic/Mystran SVF (Andrew Simper) — a delay-free loop SVF with tanh saturation:
cppfloat processSample(float x)
{
float hp = (x - 2.0f * R * v1 - g * v1 - v2)
/ (1.0f + g * (2.0f * R + g));
float bp = g * hp + v1;
float lp = g * bp + v2;

// Nonlinear state update — tanh inside the loop
v1 = tanh(g * hp + v1); // <-- this is the heart of the sound
v2 = tanh(g * bp + v2);

return hp; // or lp / bp depending on mode
}
Where g = tan(π * fc / sampleRate) and R = 1 / (2 * resonance).
Step 2: Oversampling (critical!)
Waveshapers inside the loop generate aliasing, which is why oversampling or other countermeasures are necessary. Use at least 2x oversampling, preferably 4x. In JUCE this is straightforward with dsp::Oversampling. Willpirkle
Step 3: Drive Before the Filter
The "grungy" character intensifies further if you add a drive parameter (soft-clip) before the SVF:
cppfloat driven = tanh(drive * input);
float filtered = svf.processSample(driven);
Step 4: HP/BP/LP Output Blend
Since LP, BP, and HP are all available, you can expose a mode blend parameter:
cppfloat out = lpGain * lp + bpGain * bp + hpGain * hp;

JUCE Architecture
PluginProcessor
├── Oversampler (dsp::Oversampling, 2x or 4x)
├── NonlinearSVF (custom class)
│ ├── g, R, v1, v2 as state variables
│ ├── tanh() in state update
│ └── LP / BP / HP outputs
└── Parameters: cutoff, resonance, drive, mode (LP/BP/HP/blend)
The critical difference from the standard JUCE dsp::StateVariableTPTFilter: the JUCE built-in has no nonlinearities inside the loop. You need to implement the SVF manually.

Why Does It Sound Like Ring Modulation?
When self-oscillation runs at frequency f_res and your saw comes in at f_note, the nonlinear feedback generates products at f_note ± f_res, f_note ± 2·f_res, etc. — identical to the sonic signature of a ring modulator, but emerging entirely from within the filter physics. That's why you haven't heard this sound anywhere else: it's a side effect of the nonlinearity, not a dedicated module.

Want me to write out the complete JUCE C++ implementation of the nonlinear SVF?

Moog ladder is still possible though. I would throw this post on Claude and provide the code parts you have.

EDIT, now claude thinks it could also be a ...

Huovilainen Ladder (der wahrscheinlichste Kandidat)
cpp// Jede Stage hat einen Tanh-Eingang
float processSample(float x)
{
float feedback = resonance * 4.0f * tanh(output); // nichtlineare FB

float s0 = tanh(x - feedback);
float s1 = tanh(stage[0] + g * (s0 - stage[0]));
float s2 = tanh(stage[1] + g * (s1 - stage[1]));
float s3 = tanh(stage[2] + g * (s2 - stage[2]));
float s4 = tanh(stage[3] + g * (s3 - stage[3]));

// LP = s4
// HP = s0 - s4 (oder x - s4)
// BP = s2 - s4

stage[0]=s1; stage[1]=s2; stage[2]=s3; stage[3]=s4;
output = s4;
}

gtps response...



This helps a lot. The nonlinear SVF idea makes sense with what I’m hearing.

The strongest clue in my tests is that the effect is strongest in HP, somewhat present in BP, and mostly absent in LP. Ring and sync are off, only one oscillator is active, and changing OSC B pitch does nothing. Also, with filter keytrack at 0%, each note creates a different pulse/beating speed; raising keytrack makes the pulse more consistent.

So I agree that this is probably not a separate ring modulator. It seems more like the filter resonance is acting like a second internal tone, and the nonlinear filter feedback/state path is creating intermodulation between the saw harmonics and the resonant frequency.

The part that interests me most is where the nonlinearity is placed. My attempts using external drive or a dirty resonant HP filter got too harsh and edgy, while PoiZone’s resonance is very strong but smooth and refined. So maybe the missing part is not “more distortion,” but a soft nonlinearity inside the SVF/lattice/state feedback structure, plus proper gain compensation and smoothing.

I’m going to test a few versions:

1. clean SVF HP/BP
2. nonlinear SVF with saturation inside the state/feedback path
3. input-drive SVF
4. ladder-derived HP/BP, possibly Huovilainen-style
5. oversampled versions to see whether aliasing/harshness is part of the mismatch

My instinct is that a nonlinear multimode SVF is the best candidate so far, but the ladder-derived HP/BP idea is worth A/B testing too.

[Touch The Universe]

I'm thinkig maybe the secret is in the resonance itself so I'm looking there next. Notice the ringmod affect goes away when keytrack is all the way up. this is an important clue.




a high-resonance nonlinear multimode filter where the resonance feedback/state path creates the intermodulation internally.

So the next code direction should be something like:

PZ Nonlinear SVF HP
PZ Nonlinear SVF BP
optional Ladder-derived HP/BP as a comparison
resonance nonlinearity inside the filter state/feedback path
drive default at 0
oversampling if needed
body blend / resonance smoothing
keytrack behavior carefully tuned

Your phrase “it turns on like a motor” is actually a great descriptor. That sounds like the resonance crosses a threshold where the resonant peak becomes strong enough to beat/intermodulate with the saw in a stable repeating way.



i think i need a more technical test to make use of gtp, this was random